[Cbc] Tie-breaking a model

[rhavar at protonmail.com]

I actually came up with a different idea, what do you think?

First I ask the solver to minimize obj1  .. and when it returns a solution (e.g. 200) then I get the model and add the constraint

obj1 = 200

But this time, tell it to minimize obj2 instead of obj1. And the constraint will force it to pick from where obj1 is ideal

This should help with the numerical issues, right?

-Ryan

-------- Original Message --------
On February 16, 2018 8:15 AM, Haroldo Santos <haroldo.santos at gmail.com> wrote:

> Hi,
> In general, use the smaller possible value for upperBound.
> Gurobi supports hierarchical objectives
> http://www.gurobi.com/resources/seminars-and-videos/numerical-issues-webinar
> Cbc does not have native support for hierarchical objectives but maybe you could use callbacks to interrupt whenever a new integer feasible solution is produced and evaluate this solution using a lexicographic comparison.
> Cheers,
>
> On Fri, Feb 9, 2018 at 7:41 PM, <rhavar at protonmail.com> wrote:
>
>> I routinely am running into an issue where in the model I have multiple objectives. Say obj1 and obj2. Now obj1 is a lot more important for me, but it's very frequent that ties exist. So I want to say: minimize for obj1, but if there's a tie minimize obj2.
>>
>> I can sort of do this already by computing a max-possible upper bound for obj2 (but it's very big)  and then say
>>
>> Minimize:
>>    obj1 * upperBound  + obj2
>>
>> which of course works, but I think it leads to some huge numbers that solvers struggle with. I wonder is there a smarter way of "tie breaking"? Or perhaps some sort of native feature where you can say  Minimize(obj1, obj2) ?
>>
>> -Ryan
>>
>> _______________________________________________
>> Cbc mailing list
>> Cbc at list.coin-or.org
>> https://list.coin-or.org/mailman/listinfo/cbc
>
> --
> =============================================================
> Haroldo Gambini Santos
> Computing Department
> Universidade Federal de Ouro Preto - UFOP
> email: haroldo [at ] iceb.ufop.br
> home/research page: [www.decom.ufop.br/haroldo](http://www.decom.ufop.br/haroldo/)
>
> It has long been an axiom of mine that the little things are infinitely
> the most important.
> -- Sir Arthur Conan Doyle, "A Case of Identity"
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.coin-or.org/pipermail/cbc/attachments/20180216/514abac5/attachment.html>